We can only see a short distance ahead, but we can see plenty there that needs to be done.”
"
—Alan Turing
OpenAI said its upcoming AI model Astra has shown “critical” cyber capabilities under the company’s internal framework and will take longer to develop with necessary safeguards.
Tech giant Meta has confirmed that one of its AI models accessed the internet and hacked into a third-party company’s systems during a cybersecurity test after a misconfiguration error by independent testing partner Irregular. The incident places Meta alongside OpenAI and Anthropic, whose models also recently breached external organizations while under evaluation.
AI researcher Andrej Karpathy has outlined a vision for the future of gaming after Claude Opus 5 generated 5,500 lines of Three.js code to turn the opening of The Lord of the Rings into a playable low-poly 3D world, all from a single paragraph and a $10 budget. He calls it the beginning of “ephemeral GTA of X on demand.
The launch comes two days after the US FCC announced restrictions on new foreign-made humanoid robots, citing “unacceptable risks” to US national security. Google DeepMind said Gemini Robotics 2 includes new safety features for real-world operation and human collaboration.
Days after OpenAI revealed that its autonomous AI agents escaped a testing sandbox and breached Hugging Face (and other organizations), the company has engaged independent nonprofits METR and Redwood Research to review the model behavior.
Sarvam AI announces it is building a trillion-plus parameter foundational model in India at Epoch 2026, while claiming its flagship 105B is 5.5× cheaper than GPT-5.4 Mini.
The launch of Kimi K3 has given a sense of deja vu for the American tech industry that saw markets collapsing with the release of DeepSeek in 2025.
The latest move marks Anthropic’s second consecutive extension since the model’s return earlier this month after U.S. export control restrictions temporarily halted global availability.
From rendering video game graphics to powering nearly every AI model on the planet, NVIDIA’s rise is no accident. We break down the three-decade stack of bets — GPUs, CUDA, and now open models — that made one chipmaker the most valuable company in history.
loudflare’s research on Anthropic’s Claude Mythos Preview, conducted under Project Glasswing, found the frontier model can stitch together low-severity bugs into a single, working exploit — a capability that sets it apart from other frontier models and raises new questions about safeguards on capable cyber AI.
